Dictation of multiparty conversation using MLLR speaker adaptation and statistical turn taking model

摘要

A new speech decoder dealing with multiparty conversation is proposed. Multiparty conversation denotes a situation in which many speakers talk each other. In such a situation, the system has to recognize not only the word sequence of the input speech but also the speaker of each part of them. We propose the method utilizing not only acoustic model and language model, which are the resources of conventional single-user speech decoder, but also stochastic turn taking model and speakers individual models using MLLR speaker adaptation to recognize speech. This framework realizes simultaneous maximum likelihood estimation of spoken word sequence and the speaker sequence. Experimental results using TV sports news show that the proposed method reduce the word error rate by 29.5 % and speaker error rate by 89.7 % compared to the conventional method.